LABOUR-MARKET ATTACHMENT AND TRAINING PARTICIPATION

LABOUR-MARKET ATTACHMENT ANDTRAINING PARTICIPATION*

By TOSHIE IKENAGA† and DAIJI KAWAGUCHI‡

†Cabinet Office, Government of Japan ‡Hitotsubashi University

This paper examines how expected attachment to the labour market and expected tenureat a specific firm affect training participation. The results, based on cross-sectional datafrom Japan, indicate that expected attachment to the labour market affects participation inboth employer- and worker-initiated training, while expected tenure at a specific firmmainly explains participation in employer-initiated training. These two attachment indicesexplain more than two-thirds of the sex gap in training participation. Employers inless-competitive labour markets are more likely to offer employer-initiated training totheir workers.jere_573 1..25

JEL Classification Numbers: J24, J42.

1. Introduction

Job training and skill development play a central role in the formation of job skills andsubsequent wage growth (e.g., Lynch, 1992; Kurosawa, 2001; Kawaguchi, 2006).Although the share of women and non-regular employees in the workforce has steadilyincreased in Japan, their job-training opportunities remain substantially limited whencompared to those of male and regular employees (Hara et al., 2009; Kosugi and Kimura,2009).

The lower rate of training participation among female and non-regular workers is oftenattributed to their shorter expected periods of labour-force participation or job tenure witha specific firm. Indeed, theory suggests that these expected lengths are important deter-minants for the quantity of human-capital investment, because the strength of labour-market attachment and expected job tenure determine the length of time that agents canreap returns to their human-capital investment. In particular, when human capital isfirm-specific for technological reasons or because of labour-market friction, the costsinvolved in human-capital investment is paid by firms or shared between firms andemployees (Hashimoto, 1981; Stevens, 1994; Chang and Wang, 1996; Acemoglu andPischke, 1998, 1999). Under these circumstances, to secure their investment, firms arelikely to invest more in employees whom they expect to stay at the firm for a longer time.

These predictions of human-capital theory are well known, but empirical tests of thesepredictions are scarce. One notable exception that directly tests these predictions isRoyalty (1996), who used panel data from the National Longitudinal Survey of Youth1979 (NLSY79) of the USA to estimate job-to-job and job-to-non-employment turnoverprobabilities and showed that the estimated probabilities explain well the probability of

* This paper is a revised version of Toshie Ikenaga and Daiji Kawaguchi, “Training Opportunities for‘Marginal Workers’ in Japan” PIE/CIS DP-467, Hitotsubashi University. The Statistics Bureau, Ministryof Internal Affairs and Communications granted special permission to use microdata from the Employ-ment Status Survey. We would like to thank two anonymous referees, participants in a workshop at theResearch Institute of Economy, Trade and Industry and the International Collaboration Project by theEconomic and Social Research Institute of the Cabinet Office for valuable suggestions.

bs_bs_bannerThe Japanese Economic Review

The Journal of the Japanese Economic Association

The Japanese Economic Review doi: 10.1111/j.1468-5876.2012.00573.x

1© 2012 Japanese Economic Association

receiving training. Loewenstein and Spletzer (1997) offered indirect evidence consistentwith theoretical predictions. They claimed that both employers and employees have anincentive to delay the timing of formal on-the-job training, because they postpone it untilthey learn the quality of the current employer–employee match. Once both sides learnthat the match is good and expect the relationship to last for a long period, both sides startto invest in human capital. Loewenstein and Spletzer (1997) indeed found that on-the-jobtraining tends to take place after a few years of job tenure, even after conditioning on theyears of completed job tenure to control for the quality of job match, based on theNLSY79. Brunello and Gambarotto (2007) empirically investigated the correlationbetween labour-market competition and employer-provided training and found thatemployer-provided training in the UK occurs less frequently in economically denserareas. They argued that poaching and turnover effects of agglomeration discourageemployers from providing training.

This paper proposes an alternative test of the theoretical prediction, relying on a singlecross-section of data from the Employment Status Survey 2007 under a stationaryassumption. We first calculate an “attachment index” for each worker, that is, how longeach worker is expected to stay in the labour market until retirement, by adding up theaverage hours worked until the standard retirement age for each of the worker’s attributes.In addition, we similarly calculate each worker’s expected “remaining tenure,” that is,how many more years each worker is expected to continue working at the current firm,based on workers’ observable characteristics. Greater attachment to the labour market asa whole implies a longer payoff period for investment in general human capital, and thisshould increase job training initiated by both employers and workers. At the same time,longer remaining tenure (RT) implies a longer payoff period for firm-specific human-capital investment, and this should increase employer-initiated training. We then examineto what extent differences in indices in general labour-market attachment and in specific-firm attachment explain differences in the participation rates of employer- and worker-initiated trainings by sex, education, and regular/non-regular employment status. Thecorrelation between the expected length of job tenure and employer-initiated job trainingis predicted to be stronger in labour markets with more significant labour-market friction,because firms can exploit higher rents from human-capital investment. We constructproxy variables for labour-market friction and examine how the correlation betweenexpected tenure and employer-initiated training differs by the degree of labour-marketfriction.

The main findings of our analysis are as follows. First, whereas the predicted future-employment period overall affects participation in both employer- and worker-initiatedtrainings, the predicted future employment period at a particular firm mainly affectsparticipation in employer-initiated training. Second, expected labour-market attachmentexplains more than two-thirds of the difference between men and women in the prob-ability of participating in employer-initiated training. In contrast, these proxy variablesexplain very little of the difference in training probabilities between regular and non-regular workers. These results suggest that a considerable part of the difference injob-training participation between men and women is the result of differences in theirfuture prospects for labour-market attachment, while differences in job-training partici-pation between regular and non-regular workers arise largely because of the difference inskill requirements between the two groups. Third, firms in more competitive local labourmarkets are less likely to offer employer-initiated training to their workers, presumably inthe face of higher poaching risk.

The Japanese Economic Review


It should be noted as a caveat that we cannot fully identify the causal effect of the lengthof planning period on human-capital-investment behaviours because of the usage of asingle cross-section of data for the analysis. The length of planning period is inferred partlybased on exogenous factors, such as sex, but partly based on endogenous factors, such aseducational attainment. As how long an individual receives education is a matter ofindividual choice, more able individuals may well invest more in human capital throughboth schooling and training participation. More generally speaking, individuals with higherpotential earning capacities arguably plan to stay in the labour market longer and partici-pate in training activities more vigorously. Accordingly, the estimated effect of the planningperiod on training participation in this paper is likely to be the upper bound of the truecausal effect. Having stated this limitation, our contribution to the literature is not offeringclean estimates of the causal effect, but articulating the importance of the planning horizonto establish a link between workers’ characteristics and training participation.

The remainder of this study is organized as follows. Section 2 outlines the data andpresents the heterogeneous training-participation rate across workers’ attributes. Section3 constructs measures of labour-market attachment and expected length of job tenure byworkers’ characteristics and examines the extent to which these measures can explainpatterns of training participation by workers’ characteristics. Section 4 explores theimplications of local labour-market friction on training participation. Section 5 providesconclusions.

2. Data and descriptive analysis

The source of our microdata is the 2007 Employment Status Survey, which is a householdsurvey of Japan that records employer-initiated and worker-initiated training.1 Distin-guishing between whether training was conducted at the employer’s initiative or that ofthe worker himself, the survey provides a breakdown of such training into the followingcategories: (a) training at the workplace (this category applies only to employer-initiatedtraining); (b) attending college or graduate-school courses; (c) attending courses at aspecial training school or other vocational school; (d) attending courses at a publicoccupational skills development facility; (e) attending short courses or seminars; (f)participating in study-group meetings or workshops; (g) taking distance-learning courses;(h) self-learning (this category applies only to worker-initiated); and (i) other. In thisstudy, we refer to training initiated by the employer as “employer-initiated training” andtraining initiated by the worker as “worker-initiated training”.

We limit our sample to employed persons aged 15–59 and exclude those enrolled ineducation. Moreover, we exclude company executives, the self-employed (with or withoutemployees), family workers, and those doing piecework at home, because their workstatus is somewhat different in nature from the concept of an “employee” that we focuson here. Furthermore, we exclude observations for individuals when we think there arerecording errors.2 Table 1 reports descriptive statistics of the analysis sample.

1 Employer-initiated training refers to the case in which employers offer training opportunities and usuallyfund training costs. Worker-initiated training refers to the case in which workers participate in trainingindependently.

2 For example, cases in which: the years of tenure are greater than 45, the age at which the present job wastaken up is less than 15, etc.

T. Ikenaga and D. Kawaguchi: Labour-market Attachment and Training Participation


In this analysis sample, we reconfirm the findings of previous works on the correlationbetween workers’ characteristics and training participation (Altonji and Spletzer, 1991;Green, 1993; Pischke, 2001; Kurosawa, 2001; Kawaguchi, 2006; Hara et al., 2009;Kosugi and Kimura, 2009). Male, regular, younger, educated workers are more likely toparticipate in employer-initiated training than female, non-regular, older, less-educatedworkers. Workers in larger firms are more likely to participate in employer-initiatedtraining. The difference in the probability of participating in worker-initiated training by

TABLE 1Descriptive statistics of analysis sample, employed workers aged 15–59, n = 374,468

Mean Standard deviation

Female (%) 46.4 —Regular employees (%) 69.7 —Part-time and casual workers (%) 20.9 —Dispatched workers from temporary labour agencies (%) 2.7 —Contract employees (%) 5.0 —Primary or junior high school (reference) (%) 7.0 —Senior high school (%) 46.8 —Vocational school, junior college (%) 22.8 —College, graduate school (%) 22.2 —Age (years) 40.8 11.215 to 19 (%) 1.0 —20 to 24 (%) 7.8 —25 to 29 (%) 11.1 —30 to 34 (%) 13.0 —35 to 39 (%) 13.3 —40 to 44 (%) 12.7 —45 to 49 (%) 13.3 —50 to 54 (%) 13.3 —55 to 59 (%) 14.5 —Firm size (number of employees): 1 to 9 (%) 13.8 —10 to 29 (%) 13.6 —30 to 99 (%) 15.9 —100 to 299 (%) 13.6 —300 to 499 (%) 5.6 —500 to 999 (%) 6.1 —1,000 and over (%) 18.9 —Government (%) 11.8 —Tenure (years) 11.5 10.70 to 4 (%) 38.2 —5 to 9 (%) 16.9 —10 to 14 (%) 12.0 —15 to 19 (%) 10.8 —20 to 24 (%) 6.9 —25 to 29 (%) 6.2 —30 to 34 (%) 5.0 —35 to 39 (%) 3.3 —40 and over (%) 0.8 —New graduates dummies (%) 23.5 —Attachment index 11.7 8.4Remaining tenure (RT) | RT �0 2.2 3.6Dummy for RT < 0 (%) 48.12 —Number of employees per 1,000 km2 0.28 0.47Industry specialization 0.09 0.05Average years of education 13.26 0.34Fraction of college graduates (%) 19.5 —



workers’ characteristics is not as stark as the case for employer-initiated training. A rathersurprising finding is that the participation rate increases with workers’ job tenure, with apeak of about 50 percent for those with a tenure of 25–29 years, according to the simpletabulation results. The differences of training participation probability by workers’ char-acteristics are preserved in multivariate frameworks. It is particularly notable that, in themultiple regression analysis, the probability of participating in employer-initiated trainingincreases in an almost linear fashion until 40 years of job tenure. These findings areconsistent with the notion of “continuous skill upgrading” in Japanese workplaces (Hash-imoto and Raisian, 1985; Hart and Kawasaki, 1999). For the sake of reference, AppendixTable 1 tabulates training participation probability by workers’ characteristics, andAppendix Table 2 tabulates probit regression results.

3. Correlations among labour-market attachment, RT, and job training

3.1 Measurements of planning period

Human-capital models claim that the amount of investment in general human capital at aparticular point in time is determined by the marginal rate of return on investment andmarginal cost. A key determinant of the marginal rate of return on investment is thelength of the payoff period. In addition, when human capital is firm-specific, the firmreaps part of the gap between workers’ marginal productivity and their market wage asrent, and the discounted present value of that rent determines the amount of human-capital investment financed by the firm. The discounted present value of that rent cru-cially depends on workers’ remaining employment period.

The purpose here is to examine to what extent we can explain differences in trainingprobabilities across workers’ attributes found in the preceding section with differences inworkers’ remaining employment period.

3.1.1 The attachment index

The more workers are attached to the labour market, the higher is their incentive toparticipate in training and raise their job skills. The degrees of labour-market attachmentpresumably differ by workers’ characteristics, such as age, sex, or educational back-ground. To gauge this labour-market attachment, we calculate the total amount of timeeach worker can be expected to spend in the labour market under the assumption that theworker behaves as the average person within the demographic group to which he/shebelongs.

Specifically, the attachment index (AI) is calculated as:

AI age sex education hours sex educationtt age, , ,( ) = ( )

=∑ 200059

where hours sex educationt ,( ) is the average hours worked3 by workers of t years old,defined by workers’ sex and educational backgrounds. The summed hours worked until59 years old, the general retirement age, is divided by 2,000, which is the typical numberof annual hours worked by a full-time worker. This AI index attempts to capture the total

3 We apply zero in the case of those not employed.



strength of labour-market attachment before retirement age; thus the sample now includesthose out of the labour force and employed persons who are in education, companyexecutives, the self-employed (with or without employees), family workers, and thosedoing piecework at home. The sample of 15–59-year-olds (sample A) is divided into 442groups according to their attributes (age, sex, education). Next, we divide the sample ofemployed persons used in the estimation in Section 3 (sample B) into groups accordingto the same attributes (age, sex, education) (415 groups). We then apply the AI of aparticular group in sample A to each of the same 415 groups in sample B.

This index is an indicator showing how many full-time years a worker of a given sexand with a given education will work in the period that remains from his or her age untilage 59. It should be noted that we implicitly assume a stationary economic environment,because we take the average employment patterns in the Employment Status Survey forthe observations and assume that the cross-sectional observations represent observationsof the employment patterns for individuals over time. This is a strong assumption, but itis a standard one made, for example, in estimations of Mincerian wage equations usingcross-sectional data.

3.1.2 Remaining tenure

How long a worker with given attributes is expected to continue working for the presentemployer is likely to be an important determinant of employer-initiated training. As oursecond measure, we calculate the expected RT for each attribute, which gauges how longa worker with given attributes can be expected to continue working for the presentemployer.

To calculate the expected RT period of a worker with certain demographic character-istics, we calculate the following index:

RT sex education employment status industry size of employe, , , , rr

directly hired from schools tenure sex education emplo

,

, ,

() = yyment status

industry size of employer directly hired fr,

, ,(

oom schools tenure) −

based on the sample of employed persons from sample B in the previous subsection.There are 6,151 groups according to workers’ attributes (sex, education, employmentstatus, industry, size of employer, and directly hired from schools (whether workersentered a firm directly upon graduation)).4 The variable tenure is the median years of jobtenure for each demographic group. Because the number of observations may be verysmall for some groups, we employ the median to avoid distortion from outliers. Wesubtract the actual years of tenure from the median value of years of tenure for eachgroup and set this as RT. If the value thus obtained is negative, we set RT to zero.Moreover, we create a dummy that takes a value of 1 if the value obtained is negative, torepresent strong attachment to a firm that is unascertainable from workers’ observableattributes.

The variable “directly hired from schools” indicates whether the worker took up thecurrent employment right after his/her school graduation. In the Japanese labour market,

4 We do not consider occupation as a workers’ attribute because workers’ occupation can change with age,such as when workers move into administrative and managerial occupations.



there is a strong tendency for fresh graduate recruits to follow a career path throughpromotion within the firm, while mid-career recruits represent a much more fluid workforce and can be expected to subsequently follow a career through job changes. Here, wemechanically regard as having started their present job as fresh graduate recruits those forwhom the age at which they took up the job (current age minus years of tenure) was15–16 years in the case of junior-high-school graduates; 18–19 years in the case ofhigh-school graduates; 20–21 years in the case of graduates of vocational schools, juniorcolleges, or technical colleges; and 22–25 in the case of graduates of colleges andgraduate schools.

Figure 1 shows the distribution, average, and median for the RT of 30-year-old maleregular employees who graduated from college or graduate school, with the upper panelfor fresh graduate recruits and the lower panel for mid-career recruits. Whereas the RT ofgraduate recruits is around 12 years, that for mid-career recruits, even though theyotherwise have the same attributes in terms of sex, education, and employment status, isstrikingly lower, at around 2 years. Based on this result, we expect that those recruitedupon graduation hold jobs that they will continue to work in for a long time, and theprobability that they will receive employer-initiated training is consequently high.

We should note a caveat about the RT measure here. A preferable measure as adeterminant of firm-initiated training participation would be the expected length ofcompleted tenure. Because of a lack of an appropriate longitudinal data set that includesinformation about completed tenure, we use an available single cross-section of data andconstruct the RT measure from the distribution of uncompleted tenure. The distributionsof uncompleted tenure and completed tenure are generally different. From an age-specificdistribution of uncompleted tenure of a single birth-year cohort based on repeatedcross-section data, however, the distribution of completed tenure for the birth-year cohortcan be recovered. In addition, under an assumption that job-quit behaviour does not differacross birth cohorts, the age-specific distribution of uncompleted tenure is obtained froma single cross-section of data (Hall, 1982). In sum, the expected length of completedtenure is a monotonic increasing function of the RT measure in a stationary environment.Instead of imposing the exact shape of the function that transforms RT to the expectedlength of completed tenure, we start our examination of the correlation between RT andfirm-initiated training participation in a non-parametric way without assuming a specificfunctional form.

3.2 AI, RT, and training participations

We now attempt to explain the difference in training-participation probabilities acrossdemographic groups by the difference in the expected length of labour-market attachmentor tenure at a specific firm. In particular, we examine whether the lower rate of trainingparticipation by female and non-regular workers can be explained by the shorter length ofexpected length of labour-market attachment or tenure at a specific firm. Table 2 tabulatesthe means and standard deviations of the AI and RT by demographic characteristics andemployment status. Here Table 2 focuses only on the statistics of RT with 0 and over. Thefigures indicate that female workers tend to have a lower average of both AI and RT. Alltypes of non-regular workers have a shorter expected length of RT than regular workers.

We start by looking at the effects of AI and RT on the probabilities of trainingparticipation by types of trainings. To identify the correlations, the following probitmodels are estimated:



0

0.1

0.2

0.3

0

(a)

(b)

5 10 15 20

0

0.05

0.1

0.15

0.2

−10 -5 0 5 10 15

Remaining Tenure

Den

sity

Remaining Tenure

Den

sity

FIGURE 1. Remaining tenure: 30-year-old male regular employees who have graduated from college orgraduate school. (a) Fresh graduate recruits (median = 12.000, mean = 12.126); (b) mid-career recruits(median = 2.000, mean = 2.203)



Pr , ,,Training NegativeRTi i=( ) = + + +( )1 0 1 2 3AI RT AI RTi i i iΦ β β β β

where “Trainingi” is a dummy variable indicating whether person i received employer- orworker-initiated training, and AIi and RTi are sets of dummy variables that correspond tothe years of AI or RT of person i. The dummy variable NegativeRT1 takes one if RT isnegative for person i.

Results are presented in Figures 2 and 3, which on the horizontal axis show the valuesof the dummy variables and on the vertical axis indicate the size of the marginal effectestimated from the probit estimation. As can be seen, for AI, the higher the index (i.e., thegreater the predicted future labour-market attachment), the higher the training probabilityis. There are no great differences in the shapes of the curves for employer- and worker-initiated training. For RT, we also find that the higher the value, the higher the trainingprobability is, but there is a considerable difference in the shapes of the curves for the twotypes of training. That is, whereas the probability of employer-initiated training displaysa steep increase, the probability of worker-initiated moves sideways until 8 years of RT,and after that it rises relatively slowly. This result shows that whereas a greater length offuture employment, as represented by AI, is associated with an increase in job training atthe initiative of both workers and firms, a greater length of predicted employment at aspecific firm, represented by RT, is associated mainly with an increase in job training atthe firms’ initiative. These results are consistent with human-capital theory, under theassumption that firms do not fully compensate workers for their skill upgrading inducedby training participation because of skill specificity or labour-market friction, and thusfirms have an incentive to invest in workers to reap the return to investment.

We now attempt to further decompose the contents of employer-initiated training.Among the forms of employer-initiated training are those that take place at the workplace,in college/graduate schools, or in training/vocational schools. Training that takes place inworkplaces presumably endows human capital that is specific to those particular envi-ronments, whereas training that takes place in schools endows human capital that isgeneral across workplaces. Therefore, even among the forms of employer-initiated train-ing participation, the AI or RT presumably predicts training participation at workplaces

TABLE 2Means and standard deviations of attachment index and remaining tenure by demographic characteristics,

employed workers aged 15–59

Attachment index Remaining tenure 0 and over

MeanStandarddeviation Mean

Standarddeviation

Male 14.9 9.3 5.1 4.4Female 8.1 5.1 3.2 3.1Regular employees 12.7 8.5 5.2 4.3Part-time and casual workers 8.3 6.9 2.0 1.5Dispatched workers from temporary

labour agencies13.8 7.8 0.9 0.9

Contract workers 11.4 8.5 1.9 2.0Primary or junior high school 7.9 8.1 5.1 5.1Senior high school 11.1 8.5 4.6 4.2Vocational school, junior college 12.2 7.9 3.3 3.1College, graduate school 13.9 8.3 4.1 3.9



better. For the sake of a parsimonious estimation, the indicators of training participationare regressed on linear terms of AI, RT, and an indicator for negative RT. The results ofthe estimations are reported in Table 3. Column (1) reports the regression results of allemployer-initiated training on linear terms of AI and RT. In parallel with Figures 2 and 3,both stronger labour-market attachment and longer expected RT are associated with ahigher probability of employer-initiated training participation. The positive coefficient forthe indicator for negative RT needs some explanation. The negative value of RT impliesthat those workers whose current tenure is longer than the predicted tenure have aparticularly strong attachment with the current employer based on unobserved reasons.Thus, the finding that those workers are more likely to participate in employer-initiatedtraining is not particularly surprising.

FIGURE 2. The attachment index (AI) and training probabilities.Note: Probit regression coefficients on dummy variables AIi in Pr(Trainingi = 1|AIi, RTi) = F(b0 + AIib1 +RTib2 + NegativeRTib3) are reported on the vertical axis. Marginal effects at the means of independentvariables are reported. All coefficients are statistically different from zero.



Decomposing the contents of employer-initiated training by the venues of trainings, thetraining participation that takes place at workplaces is more strongly associated with AIor RT than the training participation that takes place at schools, as the results reported inColumns (2) to (4) in Table 3 indicate. These results imply that firms choose who willparticipate in workplace training carefully, based on the workers’ expected attachment tothe labour market or the current employer. For school-based trainings, other factors, suchas general learning ability, are more significant factors for selection than the expectedplanning period. For the sake of completeness, the results for worker-initiated training arereported in Column (5). We reconfirm that the AI explains participation in worker-initiated training but RT does not explain it.

FIGURE 3. Remaining tenure (RT) and training probabilities.Note: Probit regression coefficients on dummy variables RTi in Pr(Trainingi = 1|AIi, RTi) = F(b0 + AIib1 +RTib2 + NegativeRTib3) are reported on the vertical axis. Marginal effects at the means of independent variablesare reported. The coefficients for “Employment-initiated training” are significant for RT values from 3 and up.The coefficients for “Worker-initiated training” are significant for RT values of 1, 2, 4, 6, and 9 and up.



TA

BL

E3

Det

erm

inan

tsof

trai

ning

part

icip

atio

n,pr

obit

esti

mat

ion,

empl

oyed

wor

kers

aged

15–5

9

Init

iativ

e

(1)

(2)

(3)

(4)

(5)

Em

ploy

er-i

niti

ated

trai

ning

Wor

ker-

init

iate

dtr

aini

ng

Pla

ceA

llW

orkp

lace

Col

lege

orgr

adua

tesc

hool

Tra

inin

gor

voca

tion

alsc

hool

All

Att

achm

ent

inde

x/

100.

044

0.03

00.

001

0.00

20.

022

(0.0

07)

(0.0

07)

(0.0

00)

(0.0

00)

(0.0

05)

Rem

aini

ngte

nure

/10

0.08

60.

085

0.00

1-0

.001

0.00

6(0

.014

)(0

.013

)(0

.001

)(0

.000

)(0

.011

)R

emai

ning

tenu

re<

00.

068

0.05

70.

000

-0.0

02-0

.008

(0.0

05)

(0.0

05)

(0.0

00)

(0.0

00)

(0.0

04)

Pse

udo

R2

0.00

70.

006

0.01

00.

026

0.00

4

Not

e:N

umbe

rof

obse

rvat

ions

is37

4,46

8fo

ral

les

tim

atio

ns.

Mar

gina

lef

fect

sat

the

mea

nsof

the

inde

pend

ent

vari

able

sar

ere

port

ed.

Sta

ndar

der

rors

robu

stto

clus

teri

ngof

erro

rte

rms

wit

hin

grou

ps,

defi

ned

byse

x¥

educ

atio

n¥

empl

oym

ent

stat

us¥

indu

stry

¥si

zeof

empl

oyer

¥di

rect

lyhi

red

from

scho

ol,

are

repo

rted

inpa

rent

hese

s.C

onst

ants

are

incl

uded

inal

lsp

ecifi

cati

ons,

but

the

coef

fici

ents

are

not

repo

rted

.



3.3 How much does the difference in planning period explain the differencein training participation?

The preceding results show that the length of the expected payoff period for investmentin human capital affects participation in job training. Now, we attempt to quantify howmuch these two indices can explain the difference in the probability of training partici-pation between male and female workers or between regular and non-regular workersfound in Section 2. If short expected investment-payoff periods explain why the job-training probabilities of female and non-regular workers are low, then we would expectthat by controlling for the AI and RT variables, the gap vis-à-vis the reference groupsshould shrink.

Table 4 shows the estimation result for the probabilities of employer- and worker-initiated training using sex, employment status, and education as explanatory variables.Moreover, we also include the industry, employer size, and fresh-graduate-recruitdummies used for constructing groups in the calculation of RT. This is to take intoaccount the possibility that these factors directly affect workers’ job-training probabilitythrough technological aspects of production activities and worker heterogeneity. Theresults in columns (1) and (3) do not include AI and RT, while those in columns (2) and(4) do.

The estimated coefficient of AI is positive and statistically significant as expected, butthe coefficient for RT is negative and statistically significant. This negative sign for RT is

TABLE 4Determinants of employer-initiated training participation, probit estimations, employed workers aged 15–59

Initiative

(1) (2) (3) (4)

Employer-initiated training Worker-initiated training

Female -0.035 -0.010 0.001 0.010(0.002) (0.007) (0.003) (0.003)

Part-time and casual workers -0.186 -0.185 -0.060 -0.058(0.002) (0.006) (0.004) (0.004)

Dispatched workers -0.174 -0.181 -0.005 -0.007(0.004) (0.006) (0.006) (0.006)

Contract employees -0.108 -0.110 -0.011 -0.011(0.003) (0.006) (0.005) (0.005)

Senior high school 0.074 0.057 0.074 0.069(0.004) (0.006) (0.005) (0.005)

Vocational school, junior college 0.141 0.112 0.166 0.156(0.004) (0.008) (0.006) (0.007)

College, graduate school 0.154 0.125 0.257 0.247(0.004) (0.008) (0.008) (0.008)

Attachment index / 10 — 0.036 — 0.013(0.002) (0.002)

Remaining tenure / 10 — -0.020 — 0.004(0.006) (0.004)

Remaining tenure < 0 — 0.019 — -0.016(0.003) (0.002)

Pseudo R2 0.133 0.135 0.100 0.102

Note: Number of observations is 374,468 for all estimations. Marginal effects at the means of the independentvariables are reported. Standard errors robust to clustering of error terms within groups, defined by sex ¥education ¥ employment status ¥ industry ¥ size of employer ¥ directly hired from school, are reported inparentheses. Industry, size of employer, and new graduate dummies are also included in estimations.



obtained by the fact that RT does not vary much within a group after conditioning on sex,employment status, education, industry, size of employer, and whether the worker wasdirectly hired from school; the remaining variation is highly collinear with AI. This resultimplies that separately identifying the effects of AI and RT, conditioning on sex, employ-ment status, education, industry, size of employer, and whether the worker was directlyhired from school, is difficult. Our purpose here, however, is not to identify the separateeffects of AI and RT, but rather to examine the change of coefficients for sex, employmentstatus, and education dummy variables after controlling for differences in the planninghorizon. Therefore, we rely on Column (2) of Table 4 for the following discussion.

Comparing the results for employer-initiated training, we find that in Column (1), thedifference between men and women is 3.5 percentage points, but by controlling for thelinear effect of AI and RT in Column (2), the difference shrinks to 1.0 percentage pointand becomes statistically insignificant. That is, more than two-thirds of the differencebetween men and women in the probability of receiving employer-initiated training canbe explained by the two factors of how much longer someone will continue to beemployed in the labour market (AI).5 In contrast, for non-regular workers, the differencesdo not diminish even when AI and RT are included, as reported in Column (2). Thisimplies that the non-regular workers choose not to participate in employer-initiatedtraining not simply because their expected planning period is short; rather, the differentskill requirements for non-regular jobs as compared with regular jobs seems to explainthe difference. Finally, only about one-fifth of the low training probability for the lesseducated can be explained by these factors. This suggests that while the length of theinvestment-payoff period explains some of the difference in training probabilities by levelof educational attainment, a large part of the difference is caused by differences in thereturns from job training (that is, differences in learning efficiency) and differences in thediscount rate for future earnings.

In sum, our results indicate that differences in labour-market attachment and expectedRT at the present employer affect training probabilities in a way that is consistent with thepredictions of human-capital theory. Moreover, the results show that these factors explainthe low probabilities of training participation for women and the less educated. Incontrast, the low probability of training participation among non-regular workers cannotbe explained by differences in labour-market attachment or expected length of job tenure.The gap between regular and non-regular workers is not caused simply by the differenceof labour-market or specific-firm attachments; rather, it is based on the different skillrequirements and consequent careers designed for each type of worker. This result isconsistent with Kambayashi’s (2010) finding that the classification of workers as “non-regular” at workplaces is more important than the length of contract period as a deter-minant of training participation by dividing the workers’ careers. The distinction of careerperspective with the current employer is particularly important in workplaces in Japan,where workers’ commitment is an important input for productivity improvement, asreported by Bae et al. (2011).

The determinants for participation in worker-initiated training are reported in Columns(3) and (4) in Table 4. The result in Column (4) implies that male and female workers areequally likely to be involved in worker-initiated training, whereas part-time workers areless likely to become involved in it. Moreover, more educated workers are more likely tobe involved in training activities on their own initiative. Inclusion of the AI and RT

5 When we include AI and RT separately, significant effects are found only for AI.



increases the coefficient for the female dummy variable, as reported in Column (4). Thisimplies that, holding labour-market attachment constant, female workers are 1.0 percent-age point more likely to participate in training on their own initiative. This coefficient isidentical to the coefficient reported in Column (2). Therefore, holding labour-marketattachment constant, female workers are 1.0 percentage point less likely to participate intraining initiated by employers, but they are 1.0 percentage point more likely to partici-pate in training on their own initiative. This result can be arguably interpreted as evidencethat female workers compensate for their limited training opportunities offered byemployers by participating in training programs on their own initiative. In contrast, thecoefficients for employment status or educational background dummy variables are vir-tually unchanged from Column (3). These results again support our conjecture that theskill requirements for regular and non-regular workers are different. Also, learning abilitydiffers across workers by their educational backgrounds.

4. Competition in local labour markets and training participation

The analysis in the previous section finds that an index for “remaining tenure” in aspecific firm explains participation in employer-initiated training. This correlation couldemerge when part of the return to training is captured by the firm that offers trainingopportunities to its workers. A firm can capture a part of the return when participants’outside option does not increase because of firm specificity of the accumulated skill orfriction in a local labour market. This section further explores the implication of locallabour-market friction on participation in employer-initiated training. Specifically, we firstexamine whether local labour-market friction, measured by proxy variables, increases theprobability of participating in employer-initiated training. Second, we examine whetherthe correlation between “remaining tenure” and participation in employer-initiated train-ing is stronger in markets with a higher degree of local labour-market friction.

Local labour-market friction is measured by two indexes defined at the prefecture level.The first index is the number of employees per km2, defined as z1. This index captures theease with which a worker can find another potential employer, as adopted by previousliterature (Brunello and Gambarotto, 2007; Brunello and De Paola, 2008). The secondindex is industry specialization, i.e., a share of the number of workers in a specificindustry among all workers in a prefecture. More specifically, the index for a worker in

industry k in prefecture j is defined as zE

Ekj

k kj2 =

∑. This index captures the ease with

which a worker can find another employer in the same industry as the current employer.As shown in the previous literature, part of the human capital formed on the job,including the one accrued through training participation, could well be industry-specific(Neal, 1995). If part of human capital is industry-specific, a worker in an industryagglomeration is more likely to find another employer who appreciates her skill. In fearof workers being poached, an employer in an industry agglomeration may offer feweremployer-initiated training opportunities to its workers. The higher both z1 and z2s are, themore competitive the local labour market should be, with a higher probability that aworker will be poached by another firm.

In addition to the degree of local labour-market friction, several other local labour-market conditions may affect the probability of training participation. Workers’ higherskill level in a region enhances the efficiency of human-capital accumulation (Moretti,



2004). Part of this efficiency-enhancement effect is capitalized to local land price andlocal wage (Roback, 1982). If the efficiency enhancement effect is not fully offset by anincrease in the opportunity costs of training, however, the higher average skill of regionalworkers increases the probability of training participation. To capture this local spillovereffect of human capital, prefectural-level average years of education or fraction ofcollege-educated workers is included in the specification.

To examine the effects of local labour-market characteristics on the probability oftraining participation by worker i in prefecture j, the following probit model is estimated:

Pr , ,Training z x z z z xij i j ij j i=( ) = + + + +( )1 0 1 1 2 2 3 3Φ γ γ γ γ β

where z1j is the number of workers per 1,000 km2 in prefecture j,6 z2ij is the share ofworkers in the industry that worker i works for in prefecture j, and z3j is the average yearsof education or fraction of college-educated workers in prefecture j. Vector xi includesindividual characteristics of worker i that are: female dummy, employment-type dummies,age dummies, industry dummies, occupation dummies, employer-size dummies, anddummies for years of job tenure. To capture poaching effects related to industry-specificskill more clearly, we disaggregate “manufacturing” into seven more specific subcatego-ries7 in obtaining z2ij.

Before introducing the estimation results, we discuss possible biases of the estimator.In establishing the correlation between local economic characteristics and individuallabour-market outcomes, previous studies paid careful attention to the endogeneity ofworkers’ location choice (e.g., Rauch, 1993; Glaeser and Meré, 2001). This possibleendogeneity arises mainly because of workers’ omitted ability. In these studies, research-ers argue that local economic density is likely to be positively correlated with workers’unobserved high ability because local knowledge and individual skill are likely to becomplements in the production process. Glaeser and Meré (2001) indeed find evidencethat workers with higher unobserved ability tend to reside in metropolitan areas byallowing for individual fixed effects using workers’ panel data (NLSY79 and Panel Studyof Income Dynamics). In the absence of panel data of individual workers, the estimationin this paper cannot be free from endogeneity bias. As high-ability workers are likely tolive in urban areas and vigorously participate in training activities, the estimates for g1

presented here are likely to suffer from upward bias.Table 5 reports the results of regressions. Column 1 indicates that a higher density of

workers, measured by the number of workers per 1,000 km2 and the fraction of workersin the same industry, suppresses the probability of participating in employer-initiatedtraining. The size of the coefficients is unaffected by inclusion of the regional average ofhuman capital, as reported in Columns 2 and 3. These findings are consistent with thenotion that less competition in the local labour market encourages employers to providetraining opportunities. Considering the possible upward bias for g1 as discussed above, thenegative coefficients estimated here are striking. Higher local average human capital ispositively associated with the higher probability of participating in employer-initiatedtraining, but any causal inference from this estimation result is hampered by possibleendogeneity bias in the regional human capital variable, z2ij.

6 Area data in each prefecture are obtained from the Population Census in 2005.

7 (a) Food, beverage, tobacco and feed; (b) textile, apparel, and leather products; (c) wood products,furniture, pulp, paper products, and printing; (d) chemicals; (e) metals; (f) machinery; and (g) others.



TABLE 5Probit analysis of job-training probabilities by regional characteristics, probit estimations, employed workers

aged 15–59

Initiative

(1) (2) (3) (4) (5) (6)

Employer-initiated training Worker-initiated training

Number of employees per 1,000 km2 -0.021 -0.026 -0.025 0.014 -0.002 -0.002(0.002) (0.003) (0.003) (0.001) (0.002) (0.002)

Industry specialization -0.139 -0.133 -0.134 -0.122 -0.098 -0.097(0.040) (0.040) (0.040) (0.031) (0.031) (0.031)

Average years of education — 0.009 — — 0.030 —(0.003) (0.003)

Ratio of college graduates — — 0.045 — — 0.191(0.023) (0.017)

Female -0.035 -0.035 -0.035 -0.009 -0.009 -0.009(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

Part-time and casual workers -0.167 -0.167 -0.167 -0.057 -0.057 -0.057(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

Dispatched workers from temporary labour agencies -0.165 -0.166 -0.166 -0.013 -0.014 -0.014(0.004) (0.004) (0.004) (0.004) (0.004) (0.004)

Contract employees -0.097 -0.096 -0.097 -0.012 -0.011 -0.011(0.003) (0.003) (0.003) (0.003) (0.003) (0.003)

Senior high school 0.062 0.062 0.062 0.059 0.058 0.059(0.004) (0.004) (0.004) (0.003) (0.003) (0.003)

Vocational school, junior college 0.114 0.114 0.114 0.123 0.121 0.121(0.004) (0.004) (0.004) (0.004) (0.004) (0.004)

College, graduate school 0.134 0.133 0.133 0.198 0.194 0.194(0.005) (0.005) (0.005) (0.005) (0.005) (0.005)

Age: 20 to 24 years -0.030 -0.030 -0.030 0.016 0.016 0.016(0.008) (0.008) (0.008) (0.008) (0.008) (0.008)

Age: 25 to 29 years -0.076 -0.076 -0.076 0.019 0.019 0.019(0.007) (0.007) (0.007) (0.008) (0.008) (0.008)

Age: 30 to 34 years -0.098 -0.099 -0.099 0.020 0.019 0.019(0.007) (0.007) (0.007) (0.008) (0.008) (0.008)

Age: 35 to 39 years -0.110 -0.110 -0.110 0.012 0.012 0.012(0.007) (0.007) (0.007) (0.007) (0.007) (0.007)

Age: 40 to 44 years -0.108 -0.108 -0.108 0.010 0.010 0.010(0.007) (0.007) (0.007) (0.007) (0.007) (0.007)

Age: 45 to 49 years -0.112 -0.112 -0.112 0.002 0.002 0.002(0.007) (0.007) (0.007) (0.007) (0.007) (0.007)

Age: 50 to 54 years -0.132 -0.132 -0.132 -0.013 -0.014 -0.013(0.007) (0.007) (0.007) (0.007) (0.007) (0.007)

Age: 55 to 59 years -0.154 -0.154 -0.154 -0.026 -0.027 -0.027(0.007) (0.007) (0.007) (0.007) (0.007) (0.007)

Tenure: 5 to 9 years 0.021 0.021 0.021 -0.024 -0.024 -0.024(0.003) (0.003) (0.003) (0.002) (0.002) (0.002)

Tenure: 10 to 14 years 0.035 0.035 0.035 -0.029 -0.029 -0.029(0.003) (0.003) (0.003) (0.002) (0.002) (0.002)

Tenure: 15 to 19 years 0.052 0.052 0.052 -0.025 -0.025 -0.025(0.003) (0.003) (0.003) (0.002) (0.002) (0.002)

Tenure: 20 to 24 years 0.075 0.075 0.075 -0.018 -0.018 -0.018(0.004) (0.004) (0.004) (0.003) (0.003) (0.003)

Tenure: 25 to 29 years 0.093 0.093 0.093 -0.014 -0.014 -0.014(0.004) (0.004) (0.004) (0.003) (0.003) (0.003)

Tenure: 30 to 34 years 0.095 0.095 0.095 -0.012 -0.012 -0.012(0.005) (0.005) (0.005) (0.003) (0.003) (0.003)

Tenure: 35 to 39 years 0.096 0.096 0.096 -0.005 -0.006 -0.006(0.006) (0.006) (0.006) (0.004) (0.004) (0.004)

Tenure: 40 years and over 0.081 0.080 0.080 -0.001 -0.002 -0.002(0.010) (0.010) (0.010) (0.008) (0.008) (0.008)

Industry dummies Yes Yes Yes Yes Yes YesOccupation dummies Yes Yes Yes Yes Yes Yes

Size of employer dummies Yes Yes Yes Yes Yes YesObservations 374,468 374,468 374,468 374,468 374,468 374,468Pseudo R2 0.141 0.141 0.141 0.112 0.112 0.112

Note: Marginal effects at the means of the independent variables are reported. Standard errors robust to model misspecification arereported in parentheses.



In contrast, Table 5 Column 4 shows that the higher density of workers per km2

increases the probability of participating in worker-initiated training, while a higher shareof workers in the current employer’s industry decreases it. Even when includingprefecture-level average years of education or the fraction of college graduates, localdensity does not significantly suppress the probability of participating in worker-initiatedtraining, but the share of workers in the industry of the current employer decreases it.Less friction in the local labour market, represented by a higher density of workers, doesnot discourage worker-initiated training, but lower local specialization in the currentemployer’s industry, meaning a differentiated local labour market, encourages worker-initiated training. These results are sensible if higher possibilities for workers to move toanother industry encourage workers to be more involved in worker-initiated training andto accumulate skills not necessarily specific to the industry in which workers areemployed.

We acknowledge the endogeneity of proxy variables for local economic activity, z1j, butmere estimation bias does not explain the difference in the estimation results foremployer-initiated training and worker-initiated training. The higher the density of popu-lation, the lower the probability of participation in employer-initiated training, but nosystematic correlation is found between population density and participation in worker-initiated training.

As we found in the preceding section, the higher the AI or RT, the higher theprobability of employer-initiated training. Then we examine how this correlation differsby the degree of local labour-market friction. We divide the sample into two areas. Thearea is defined as dense if the ratio of workers in the same industry each worker faces,

zE

Ekj

k kj2 =

∑, is the median value of the whole sample or more. The area is defined as

sparse if otherwise. Figures 4 and 5 compare the correlation between AI and RT acrossdense and sparse areas. Both figures indicate that the probability of training in the sparsearea, i.e., the area of a less competitive local labour market, has a steeper slope, implyingthat the positive correlation between AI or RT and employer-initiated training is strongerin areas where the local labour market is supposed to be more frictional. The effects ofRT on employer-initiated training are especially stronger in sparse areas than in denseareas.

Overall, results for employer- and worker-initiated training and the difference in resultsfor the two types of training activities do not refute the hypothesis that employersoperating in a local labour market with high friction are more likely to offer employer-initiated training to their workers because they can reap part of the return to workers’ skillaccumulation without the fear of their workers being poached.

5. Conclusion

Using microdata from the 2007 Employment Status Survey, this study empirically exam-ined determinants of workers’ participation in employer- and worker-initiated training. Bycalculating each worker’s expected labour-market attachment — that is, how much timethat worker will spend in the labour market until retirement — and each worker’sRT — that is, how many years each worker with given attributes will continue to work forhis/her present employer — we examined the correlation of these variables with trainingparticipation. We particularly focused on the low participation probabilities for women,



the less educated, and non-regular workers and examined the extent to which expectedlabour-market attachment and RT explain these workers’ low training probabilities.

Our main findings were as follows. First, we estimated how training participationdepends on workers’ attachment to the labour market, represented by the AI, and howlong a worker can be expected to continue working for his current employer, representedby RT. The results indicated that the higher the AI (i.e., the greater the predicted futurelabour-market attachment), the higher the training probabilities are. In addition, therewere no substantial differences in the shapes of the curves for employer- and worker-initiated training. We also found that the higher the value of RT, the higher the trainingis likely to be, but the slope of the curve showing the effect of RT was much greater foremployer-initiated training than for worker-initiated training. This shows that whereasgreater length of future employment increases job-training participation at the initiative of

FIGURE 4. Labour-market attachment and employer-initiated training by industry density.Note: See Figure 2. All coefficients are statistically different from zero. AI, attachment index.



both workers and employers, differences in the predicted years of employment at aspecific firm raise job-training participation mainly with the firm’s initiative. Moreover,these results suggest that there is firm-specificity in the formation of skills throughemployer-initiated training because of technology-related factors and/or market friction.

Second, women’s lower participation rate in employer-initiated training is largelyexplained by AI in the estimation. In contrast, for non-regular workers, the negativecoefficient remains largely unchanged, even when controlling for AI. These results implythat the difference in training participation between men and women is explained by thedifference in their future prospects of labour-market attachment, while the differencebetween regular and non-regular workers is not explained by this factor. This suggeststhat the difference in careers designed for regular and non-regular workers is not basedsimply on expected attachments to the labour market or a specific firm.

FIGURE 5. Remaining tenure (RT) and employer-initiated training by industry density.Note: See Figure 3. The coefficients for “RT-dense” are statistically different from zero for RT values of 2, 4,5, and 8 and up. The coefficients for “RT-sparse” are significant for RT values from 2 and up.



Third, workers in more competitive local labour markets are less likely to participate inemployer-initiated training, conceivably because of higher poaching risk. This fact isconsistent with the notion that part of human capital formed by firm-initiated training isfirm-specific, originating from market friction, and that firms can reap the return to theirhuman-capital investment. Higher average human capital in a region is found to bepositively correlated with the workers’ active participation in both employer-initiatedtraining and especially worker-initiated training.

Overall, the results obtained in this paper are consistent with the prediction of thestandard human-capital theory: that the investment-planning horizon plays a crucial rolein investment decisions. Moreover, firms’ expectations about whether they can reap thereturns to human-capital investment are shown to be a crucial determinant for firm-initiated training. This result is consistent with predictions from a strand of literatureon who finances on-the-job training (Hashimoto, 1981; Stevens, 1994; Chang and Wang,1996; Acemoglu and Pischke, 1998, 1999).

The results obtained in this paper imply that institutional practices that enable womento stay in the labour market or a specific firm for longer periods, such as work–lifebalance policies, would at the same time enhance women’s training participation. Gov-ernment policies that encourage firms to adopt such practices may well contribute tonarrowing the gap of human-capital formation between men and women and conse-quently contribute to narrowing the sex wage gap. In contrast to the clear implication forwomen, results in this paper do not illustrate the reasons behind the low training-participation rate among non-regular workers. The possible reasons for lower participa-tion may be rigid labour-market institutions that prevent a transition from non-regular toregular jobs. Non-regular workers may conceivably be confined to dead-end jobs withouta chance to upgrade their careers, resulting in a lower return to human-capital investment.Shedding more light on the reasons for the lower training-participation rate amongnon-regular workers is left for future research.

Appendix Table 1: Job-Training Participation by Workers’Characteristics (%)

Any jobtraining

Employer-initiatedtraining

Worker-initiatedtraining

Total 41.7 33.6 20.1Sex

Male 44.8 37.1 20.2Female 37.7 29.3 19.9

Employment statusRegular employees 47.9 40.3 22.5Part-time and casual workers 22.5 15.1 11.5Dispatched workers from temporary labour

agencies29.6 16.9 17.9

Contract employees 40.6 29.1 21.7Education

Primary or junior high school 17.9 14.9 5.5Senior high school 32.1 26.8 11.8Vocational school, junior college 45.4 36.0 23.2College, graduate school 59.3 47.0 33.9

Age (years)15 to 19 36.7 32.0 11.320 to 24 48.2 39.5 23.1



Appendix Table 1: (continued)

Any jobtraining

Employer-initiatedtraining

Worker-initiatedtraining

25 to 29 47.2 36.7 25.430 to 34 43.9 34.2 22.735 to 39 41.5 32.7 20.440 to 44 41.8 33.9 20.245 to 49 42.2 35.0 19.850 to 54 38.2 32.0 16.655 to 59 32.1 26.9 13.2

IndustryAgriculture, forestry, and fisheries 21.7 13.7 11.8Mining, construction 35.2 27.8 15.2Manufacturing 34.4 28.6 13.3Electricity, gas, heat supply, and water 63.6 55.5 28.3Information and communications 52.4 38.7 30.9Transport 28.5 23.9 9.9Wholesale and retail trade 33.1 26.5 13.9Finance and insurance 62.9 55.8 27.8Real estate 44.1 31.2 25.7Eating and drinking places, accommodations 23.6 15.4 12.4Medical, health care, and welfare 59.1 49.2 33.2Education, learning support 69.3 56.6 43.6Compound services 58.9 54.2 20.5Services not elsewhere classified 40.3 30.2 20.9Government not elsewhere classified 58.3 49.7 27.5

OccupationSpecialist and technical workers 66.3 54.2 40.6Administrative and managerial workers 65.8 60.0 27.6Clerical workers 42.8 33.1 21.3Sales workers 41.0 34.3 16.8Service workers 37.8 29.0 18.8Security workers 57.8 49.5 25.0Agriculture, forestry, and fishery workers 24.5 15.8 13.4Transport and communication workers 25.9 22.2 7.9Production process and related workers 28.9 23.9 10.2

Size of employer (number of employees)1 to 9 25.2 15.3 14.610 to 29 29.4 21.3 14.730 to 99 33.9 25.9 16.1100 to 299 40.4 32.9 18.4300 to 499 44.7 36.7 20.3500 to 999 47.2 39.7 21.11,000 and over 51.1 43.9 22.6Government 64.3 55.9 34.9

Tenure (years)0 to 4 38.7 28.6 20.75 to 9 39.5 31.9 19.210 to 14 41.2 34.4 18.715 to19 44.8 38.1 19.820 to 24 48.8 42.5 22.025 to 29 50.9 45.5 21.930 to 34 49.5 44.0 20.635 to 39 44.1 39.5 16.040 and over 35.1 31.1 10.8

Source: Authors’ calculation based on data from the 2007 Employment Status Survey, Ministry of InternalAffairs and Communications.



Appendix Table 2: Determinants of Job-Training Participations, EmployedWorkers, Ages 15–59

Initiative

(1) (2) (3) (4) (5) (6) (7) (8)

Employer Worker

Female 0.024 -0.036 -0.014 -0.037 0.035 0.000 0.000 -0.009(0.002) (0.002) (0.002) (0.002) (0.001) (0.002) (0.002) (0.002)

Part-time and casualworkers

-0.193 -0.171 -0.185 -0.168 -0.075 -0.062 -0.064 -0.056(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

Dispatched workers -0.199 -0.171 -0.176 -0.163 -0.044 -0.020 -0.018 -0.012(0.003) (0.004) (0.004) (0.004) (0.003) (0.004) (0.004) (0.004)

Contract employees -0.102 -0.100 -0.096 -0.096 -0.016 -0.017 -0.010 -0.012(0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003)

Senior high school 0.081 0.070 0.064 0.065 0.074 0.067 0.059 0.059(0.004) (0.004) (0.004) (0.004) (0.003) (0.003) (0.003) (0.003)

Vocational school,junior college

0.195 0.133 0.130 0.115 0.199 0.150 0.133 0.124(0.004) (0.004) (0.004) (0.004) (0.005) (0.004) (0.004) (0.004)

College, graduateschool

0.213 0.160 0.135 0.132 0.288 0.242 0.207 0.201(0.004) (0.004) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005)

Age: 20 to 24 years -0.018 -0.031 -0.018 -0.030 0.023 0.016 0.022 0.016(0.008) (0.008) (0.008) (0.008) (0.008) (0.008) (0.008) (0.008)

Age: 25 to 29 years -0.067 -0.078 -0.065 -0.077 0.025 0.019 0.025 0.019(0.008) (0.007) (0.008) (0.007) (0.008) (0.008) (0.008) (0.008)

Age: 30 to 34 years -0.090 -0.101 -0.087 -0.099 0.025 0.020 0.025 0.020(0.007) (0.007) (0.007) (0.007) (0.008) (0.008) (0.008) (0.008)

Age: 35 to 39 years -0.099 -0.113 -0.098 -0.111 0.020 0.013 0.019 0.013(0.007) (0.007) (0.007) (0.007) (0.008) (0.007) (0.008) (0.007)

Age: 40 to 44 years -0.092 -0.110 -0.093 -0.109 0.021 0.011 0.019 0.011(0.007) (0.007) (0.007) (0.007) (0.008) (0.007) (0.008) (0.007)

Age: 45 to 49 years -0.093 -0.115 -0.092 -0.112 0.012 0.001 0.012 0.002(0.007) (0.007) (0.007) (0.007) (0.008) (0.007) (0.007) (0.007)

Age: 50 to 54 years -0.117 -0.136 -0.114 -0.133 -0.007 -0.017 -0.004 -0.013(0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007)

Age: 55 to 59 years -0.141 -0.158 -0.138 -0.156 -0.021 -0.030 -0.018 -0.026(0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007)

Size: 10 to 29 persons 0.090 0.084 0.085 0.082 -0.003 -0.005 -0.006 -0.006(0.004) (0.004) (0.004) (0.004) (0.002) (0.002) (0.002) (0.002)

Size: 30 to 99 persons 0.140 0.136 0.135 0.134 0.002 -0.001 -0.002 -0.001(0.003) (0.004) (0.004) (0.004) (0.002) (0.002) (0.002) (0.002)

Size: 100 to 299 persons 0.201 0.200 0.195 0.198 0.009 0.008 0.004 0.006(0.004) (0.004) (0.004) (0.004) (0.003) (0.003) (0.002) (0.003)

Size: 300 to 499 persons 0.243 0.244 0.238 0.241 0.023 0.026 0.018 0.023(0.004) (0.005) (0.005) (0.005) (0.003) (0.003) (0.003) (0.003)

Size: 500 to 999 persons 0.258 0.266 0.255 0.262 0.026 0.035 0.022 0.030(0.004) (0.005) (0.004) (0.005) (0.003) (0.003) (0.003) (0.003)

Size: 1,000 persons andover

0.300 0.314 0.304 0.309 0.043 0.060 0.046 0.054(0.003) (0.004) (0.003) (0.004) (0.003) (0.003) (0.003) (0.003)

Size: Government 0.362 0.287 0.316 0.278 0.118 0.047 0.065 0.037(0.004) (0.005) (0.004) (0.005) (0.003) (0.004) (0.003) (0.004)

Tenure: 5 to 9 years 0.020 0.022 0.020 0.021 -0.024 -0.023 -0.025 -0.024(0.003) (0.003) (0.003) (0.003) (0.002) (0.002) (0.002) (0.002)

Tenure: 10 to 14 years 0.029 0.036 0.032 0.035 -0.032 -0.028 -0.031 -0.029(0.003) (0.003) (0.003) (0.003) (0.002) (0.002) (0.002) (0.002)

Tenure: 15 to 19 years 0.042 0.055 0.042 0.052 -0.028 -0.021 -0.030 -0.025(0.003) (0.003) (0.003) (0.003) (0.002) (0.002) (0.002) (0.002)

Tenure: 20 to 24 years 0.070 0.081 0.066 0.076 -0.017 -0.011 -0.023 -0.018(0.004) (0.004) (0.004) (0.004) (0.003) (0.003) (0.003) (0.003)

Tenure: 25 to 29 years 0.093 0.100 0.086 0.094 -0.010 -0.007 -0.019 -0.015(0.004) (0.004) (0.004) (0.004) (0.003) (0.003) (0.003) (0.003)



Appendix Table 2: (continued)

Initiative

(1) (2) (3) (4) (5) (6) (7) (8)

Employer Worker

Tenure: 30 to 34 years 0.099 0.105 0.089 0.096 -0.005 -0.002 -0.016 -0.012(0.005) (0.005) (0.005) (0.005) (0.003) (0.003) (0.003) (0.003)

Tenure: 35 to 39 years 0.098 0.108 0.085 0.098 0.002 0.007 -0.010 -0.005(0.006) (0.006) (0.006) (0.006) (0.004) (0.004) (0.004) (0.004)

Tenure: 40 years and over 0.084 0.092 0.071 0.083 0.006 0.010 -0.005 -0.001(0.010) (0.010) (0.010) (0.010) (0.008) (0.008) (0.008) (0.008)

Industry dummies No Yes No Yes No Yes No YesOccupation dummies No No Yes Yes No No Yes YesObservations 374,468 374,468 374,468 374,468 374,468 374,468 374,468 374,468Pseudo R2 0.109 0.135 0.125 0.140 0.080 0.103 0.104 0.111

Note: Marginal effects at the means of the independent variables are reported. Standard errors robust to modelmisspecification are reported in parentheses.

Final version accepted 6 February 2012.

REFERENCES

Acemoglu, D. and J.-S. Pischke (1998) “Why Do Firms Train? Theory and Evidence”, Quarterly Journal ofEconomics, Vol. 113, No. 10, pp. 70–119.

—— and —— (1999) “The Structure of Wages and Investment in General Training”, Journal of PoliticalEconomy, Vol. 107, No. 3, pp. 539–572.

Altonji, J. G. and J. R. Spletzer (1991) “Worker Characteristics, Job Characteristics and the Receipt ofOn-the-Job Training”, Industrial and Labor Relations Review, Vol. 45, No. 1, pp. 58–79.

Bae, K.-S., T. K. Hiroyuki Chuma, D.-B. Kim and I. Ohashi (2011) “High Performance Work Practices andEmployee Voice: A Comparison of Japanese and Korean Workers”, Industrial Relations, Vol. 50, No. 1,pp. 1–29.

Brunello, G. and M. De Paola (2008) “Training and Economic Density: Some Evidence from Italian Prov-inces”, Labor Economics, Vol. 15, No. 1, pp. 118–140.

—— and F. Gambarotto (2007) “Do Spatial Agglomeration and Local Labor Market Competition AffectEmployer-Provided Training? Evidence from the UK”, Regional Science and Urban Economics, Vol. 37,No. 1, pp. 1–21.

Chang, C. and Y. Wang (1996) “Human Capital Investment under Asymmetric Information: The PigovianConjecture Revisited”, Journal of Labor Economics, Vol. 14, No. 3, pp. 505–519.

Glaeser, E. L. and D. C. Meré (2001) “Cities and Skills”, Journal of Labor Economics, Vol. 19, No. 2,pp. 316–342.

Green, F. (1993) “The Determinants of Training of Male and Female Employees in Britain”, Oxford Bulletinof Economics and Statistics, Vol. 55, No. 1, pp. 103–122.

Hall, R. (1982) “The Importance of Lifetime Jobs in the U.S. Economy”, American Economic Review, Vol. 72,No. 4, pp. 716–724.

Hara, H., M. Kurosawa and Y. Yamamoto (2009) “Hiseishain No Kigyonai Kunren Ni Tsuite No Bunseki[Analysis of Firm Provided Training of Non-regular Workers in Japan]”, JILPT Labor Policy ResearchReport No. 110.

Hart, R. A. and S. Kawasaki (1999) Work and Pay in Japan, Cambridge: Cambridge University Press.Hashimoto, M. (1981) “Firm Specific Human Capital as a Shared Investment”, American Economic Review,

Vol. 71, No. 3, pp. 475–482.—— and J. Raisian (1985) “Employment Tenure and Earnings Profiles in Japan and the U.S.”, American

Economic Review, Vol. 75, No. 4, pp. 721–735.Kambayashi, R. (2010) “Jouyou Hiseiki Roudou No Shosou [Aspects of Permanent Non-regular Workers]”,

Global COE Hi-Stat Discussion Paper Series 120, Hitotsubashi University.Kawaguchi, D. (2006) “The Incidence and Effect of Job Training among Japanese Women”, Industrial

Relations, Vol. 45, No. 3, pp. 469–477.



Kosugi, R. and Y. Kimura (2009) “Jakunensha No Shugyo Jokyo, Kyaria, Shokugyo Noryoku Kaihatsu NoGenjo [The Employment Situation, Career, and Job Training of the Young]”, JILPT Shiryo Series No. 61.

Kurosawa, M. (2001) “The Extent and Impact of Enterprise Training: The Case of Kitakyushu City”, JapaneseEconomic Review, Vol. 52, No. 2, pp. 224–241.

Loewenstein, M. A. and J. Spletzer (1997) “Delayed Formal On-the-Job Training”, Industrial and LaborRelations Review, Vol. 51, No. 1, pp. 82–99.

Lynch, L. (1992) “Private Sector Training and the Earnings of Young Workers”, American Economic Review,Vol. 82, No. 1, pp. 299–312.

Moretti, E. (2004) “Human Capital Externalities in Cities”, in J. B. Henderson and J. F. Thisse, eds, Handbookof Regional and Urban Economics, Vol. 4, Amsterdam: Elsevier, pp. 2243–2292.

Neal, D. (1995) “Industry-Specific Human Capital: Evidence from Displaced Workers”, Journal of LaborEconomics, Vol. 13, No. 4, pp. 653–677.

Pischke, J.-S. (2001) “Continuous Training in Germany”, Journal of Population Economics, Vol. 14, No. 3,pp. 523–548.

Rauch, J. E. (1993) “Productivity Gains from Geographic Concentration of Human Capital: Evidence from theCities”, Journal of Urban Economics, Vol. 34, No. 3, pp. 3–33.

Roback, J. (1982) “Wages, Rents, and the Quality of Life”, Journal of Political Economy, Vol. 90, No. 6,pp. 1257–1278.

Royalty, A. B. (1996) “The Effects of Job Turnover on the Training of Men and Women”, Industrial and LaborRelations Review, Vol. 49, No. 3, pp. 506–521.

Stevens, M. (1994) “A Theoretical Model of On-the-Job Training with Imperfect Competition”, OxfordEconomic Papers, Vol. 46, No. 4, pp. 537–562.



Documents

LABOUR-MARKET ATTACHMENT AND TRAINING PARTICIPATION